.png)
AI Copyright’s Biggest Question Is Heading for a Court Test: Does Training Count as Fair Use?
The argument over AI and books is moving toward one of its most important legal tests yet.
On 4 September, the Authors Guild and co-plaintiffs asked a federal court in New York for summary judgement in their copyright case against OpenAI and Microsoft. Their argument goes beyond the familiar question of whether AI can learn from copyrighted books. They say the companies infringed copyright by obtaining, copying and using authors’ works for training, and they want the court to reject the defence that this is fair use.
Reuters reported on 8 September that the consolidated New York litigation is shaping up as a major test of how fair use applies to generative-AI training. OpenAI and Microsoft argue that training is transformative: the systems analyse patterns rather than simply replacing the original works. The authors and publishers argue that using protected work without permission to build commercial systems harms creators and the markets around their work.
Quick Shortcuts
The question is no longer simply whether AI can read a book
For writers, this is where the debate becomes much more concrete. We are not talking about whether a machine can encounter ideas, styles or language in the abstract. We are talking about how commercial training datasets are assembled, what is copied into them, where those copies came from, and whether the resulting use falls within copyright law.
That distinction matters. A ruling that AI training can sometimes be transformative does not automatically answer every question about how the source material was obtained. Likewise, proving that a model rarely reproduces long passages of a book does not necessarily settle whether the copying involved in building the model was lawful in the first place. Those are related questions, but they are not identical ones.
Why authors should care now
The outcome could influence the economics of AI training for years. If broad training use is confirmed as fair use, technology companies may have much more freedom to train on copyrighted material without negotiating individual licences. If the court draws a narrower line, licensing, provenance and compensation could become far more important parts of the AI business model.
For authors, that affects more than potential payments. It affects whether we have meaningful control over how our work enters commercial systems. It affects publishers negotiating AI clauses. It affects independent authors who may own far more of their rights directly. And it affects the basic expectation that a creative work has a traceable relationship with the person who made it.
A book doesn’t stop belonging to its author simply because it becomes useful to a machine.
AI-assisted is not AI-authored
There is another reason I think writers need to resist turning this into a simple pro-AI versus anti-AI argument. I use AI around my work. It can help with research, visualisation, marketing, organisation and the increasingly complicated business of being an independent creative. I do not use it to replace the act of writing my books.
Those positions are entirely compatible. We can recognise useful technology while still asking where its training material came from. We can use AI as a tool while insisting that authorship, ownership and responsibility remain meaningful. And we can support innovation without accepting that every creative work on the internet automatically becomes free raw material for somebody else’s commercial system.
The real issue is provenance
One of the most useful principles emerging from the AI debate is provenance: knowing where material came from and what rights travelled with it. Writers are increasingly being asked to disclose whether AI touched their work. Publishers are experimenting with AI policies. Readers are asking what is human-made. If creators are expected to provide that level of transparency, it is reasonable to ask for comparable transparency from the companies training the models.
That does not mean every training dispute will end in the same answer. Copyright law is full of context, exceptions and competing interests. But the New York litigation could move the argument away from slogans and toward something much more useful: a clearer legal boundary between learning from culture and commercially exploiting protected work.
What happens next matters far beyond this case
A summary-judgement ruling could resolve important issues before trial, although the court could also decide that disputed facts still need to be tested. Either way, authors should watch this closely. The legal rules being shaped now may eventually determine whether licensing becomes normal, whether training-data records become expected, and how much bargaining power writers retain when their work is valuable to AI systems.
For me, the question is straightforward even if the law is not: if writers are expected to explain how AI touched our work, AI companies should be expected to explain how our work touched their models.
What do you think?
If training an AI on copyrighted books is ultimately judged to be fair use, should it still matter how those books were obtained and should authors have a right to know when their work was used?
About the Author
Rob Frankson is a science-fiction author and the creator of 121 Minutes and the Near Galaxy Saga. AI & The Author explores how artificial intelligence is changing writing, publishing, authorship and the creative business surrounding books without confusing AI assistance with AI authorship.
AI & Editorial Transparency
AI & The Author is edited and published by Rob Frankson. Artificial intelligence is used to assist with news research, initial drafting, content organisation and supporting imagery. All articles are reviewed and, where necessary, edited by Rob Frankson before publication. The opinions, editorial position and final decision to publish remain the author's.
Comments